A multi-objective cuckoo search algorithm using generalized Lèvy flight and dissimilar egg identification for multispectral image thresholding

Cuckoo Search (CS) stands as a highly efficient meta-heuristic optimization algorithm. Existing literature showcases the ability of CS in multi-objective scenarios, delineated by the three fundamental rules. However, the first rule of the algorithm incurs the cost of a generation to update a single nest, while the third rule necessitates hit-and-trial methods for parameter adjustment. To address these concerns, a multi-objective cuckoo search algorithm is proposed in this paper. The algorithm builds upon a generalized concept of Lèvy Flight for generating new solutions. Problem-specific, constraint-based strategies for identifying the best nest and dissimilar eggs are also introduced. The algorithm is further applied to solve multispectral remote sensing image thresholding problem. Prior studies have underscored the efficiency of entropy and clustering-based thresholding methods over other techniques. Nevertheless, most entropy-based approaches entail converting color images to grayscale before segmentation, potentially sacrificing crucial spectral information and consequently degrading segmentation algorithm's performance. To avoid these limitations, this research introduces an entropy-based thresholding method to segment a color image without converting it to grayscale. The experiments are carried out using very high resolution (VHR) and coarse resolution (CR) multispectral (MS) images from the satellite sensors Pl'eidas-1B and Sentinel-2b, respectively. The proposed methods undergo validation against four state-of-the-art techniques on benchmark functions and six clustering indexes, respectively.